SentiSight.ai is a web-based platform, now available in both free and paid versions, that can be used for image labeling and for developing AI-based image recognition applications for things like moderating user content, tagging and classifying images, detecting damaged products and a host of other uses.
The new version offers improved image annotation, pre-trained models and similarity searches for users of the free version and even more powerful features for paid users, including: object detection model training, downloading of offline models and the ability for labeling projects to be shared, with time-tracking for each user.
To learn more about SentiSight.ai, we conducted an interview with Karolis Uziela, PhD, SentiSight.ai team lead at Neurotechnology.
1. Can you tell us more about SentiSight.ai?
SentiSight.ai is a web-based image labeling and recognition platform. We have a powerful and convenient image annotation tool that you can use to prepare your images for deep learning model training. The image annotations can be downloaded as .json files or they can be used to train the deep learning models on-site. These models can be used via our REST API or downloaded locally and used on-premise. We also have some pre-trained models and an image similarity search algorithm that can be used out-of-the-box without any additional training.
We believe that SentiSight.ai has many potential applications in the medical field, especially for image annotation purposes. For example, we have already had clients who were using SentiSight.ai to label echocardiography images.
2. You have released a new version of SentiSight.ai. How does it differ from the previous one?
Firstly, the previous version did not have an option to download deep learning models locally, which is very useful for clients who don’t want to be dependent on their internet connection. Secondly, the previous version only had the functionality to train custom image classification models, whereas, the new version also allows users to train custom object detection models. Lastly, the new version has a shared labeling project support, where the project supervisor can see other users’ labeling times and track their progress.
3. What are the main goals of SentiSight.ai?
SentiSight.ai has two major goals. The first goal is to make the image labeling process as fast, scalable and convenient as possible. The second is to make training and using deep learning models on images as easy and intuitive as possible even for people who do not have prior experience with AI methods.
4. Do users have to always be connected to the internet to be able to use the models SentiSight.ai provides?
No, it is possible to download those models and use them locally without the internet connection. The offline version of the SentiSight.ai model works as a local REST API server to which the queries can be sent either from the same machine on which the server is running or from any other connected device, including mobile phones.
5. How does SentiSight.ai help in improving and speeding the image labeling process?
Firstly, we have many different tools and key shortcuts to assist the labeling process. For example, label names can be selected by using number keys and their correspondence to label names is fully customizable. Secondly, we have a smart labeling tool that can greatly increase the speed of bitmap labeling. Lastly, it is possible to assign predictions from deep learning models as image labels. At the moment, to do this it is required to download the predictions and re-upload them as labels, but we are planning to make it possible to review and accept AI predictions as labels on-site very soon.
6. What kind of pre-trained models does SentiSight.ai provide?
At the moment, we provide pre-trained models for general image classification and object detection, place identification and content moderation purposes.
7. What tasks can users use SentiSight.ai’s similarity search feature for?
The purpose of the similarity search feature is to find similar images in your data set to the query image. For example, it could be used to find similar animals, plants, cars or any type of commercial products. Additionally, we have a NvN similarity search option that compares all images against each other. This could, for example, be used for image de-duplication.
8. Do you utilize deep learning models for the object detection feature?
Yes, object detection, just like most of our other models is based on deep learning.
9. Can more than one person work on the same project? Can time be tracked for each user?
Yes, SentiSight.ai supports shared projects in which several users can label images at the same time. Labeling time for each user can be tracked individually and grouped by project and selected time range. User roles and permissions can also be managed by the project supervisor.
10. Where do you see SentiSight.ai in the next 5-years?
I believe that AI and image recognition are becoming increasingly important in many different areas. Therefore, creating tools that make these new technologies easily accessible to the general public is very important. We aim to be the leading platform for that.
Digital Health Buzz! aims to be the destination of choice when it comes to what’s happening in the digital health world. We are not about news and views, but informative articles and thoughts to apply in your business.